Efficacy of a sepsis clinical decision support system in identifying patients with sepsis in the emergency department

Background: Early prediction of sepsis onset is crucial for reducing mortality and the overall cost burden of sepsis treatment. Currently, few effective and accurate prediction tools are available for sepsis. Hence, in this study, we developed an effective sepsis clinical decision support system (S-CDSS) to assist emergency physicians to predict sepsis. Methods: This study included patients who had visited the emergency department (ED) of Taipei Tzu Chi Hospital, Taiwan, between January 1, 2020, and June 31, 2022. The patients were divided into a derivation cohort (n = 70,758) and a validation cohort (n = 27,545). The derivation cohort was subjected to 6-fold stratified cross-validation, reserving 20% of the data (n = 11,793) for model testing. The primary study outcome was a sepsis prediction ( International Classification of Diseases , Tenth Revision , Clinical Modification ) before discharge from the ED. The S-CDSS incorporated the LightGBM algorithm to ensure timely and accurate prediction of sepsis. The validation cohort was subjected to multivariate logistic regression to identify the associations of S-CDSS-based high- and medium-risk alerts with clinical outcomes in the overall patient cohort. For each clinical outcome in high- and medium-risk patients, we calculated the sensitivity, specificity, positive and negative predictive values, positive and negative likelihood ratios, and accuracy of S-CDSS-based predictions. Results: The S-CDSS was integrated into our hospital information system. The system featured three risk warning labels (red, yellow, and white, indicating high, medium, and low risks, respectively) to alert emergency physicians. The sensitivity and specificity of the S-CDSS in the derivation cohort were 86.9% and 92.5%, respectively. In the validation cohort, high- and medium-risk alerts were significantly associated with all clinical outcomes, exhibiting high prediction specificity for intubation, general ward admission, intensive care unit admission, ED mortality, and in-hospital mortality (93.29%, 97.32%, 94.03%, 93.04%, and 93.97%, respectively). Conclusion: Our findings suggest that the S-CDSS can effectively identify patients with suspected sepsis in the ED. Furthermore, S-CDSS-based predictions appear to be strongly associated with clinical outcomes in patients with sepsis.